# PRD: Taste Board — Visual Intelligence System **Version:** 0.1 | **Owner:** Kitt | **Date:** 2026-02-27 **Status:** DRAFT — Awaiting Assaf review --- ## 1. What Is This? An internal knowledge base that turns visual inspiration into reusable design intelligence. Not a gallery. Not a mood board. A **work tool** that makes every project faster because the team has already studied what works, why it works, and when to apply it. **One-line:** Scout → Critique → Extract → Apply → Evolve. --- ## 2. The Problem Today's workflow: - Assaf saves inspiration to mymind (421 entries) and Notion (890 entries) - When a project starts, everyone googles from scratch - No shared language for "why this works" - Past CE work isn't systematically reviewed for learnings - Visual standards exist but aren't connected to reference examples - Knowledge lives in Assaf's head, not in the system **Cost of doing nothing:** Every new project starts cold. The team rebuilds taste from zero instead of compounding. --- ## 3. Users & Use Cases ### Primary Users | User | Role | Key Action | |------|------|------------| | **Assaf** | Curator-in-chief | Saves, qualifies, approves entries into knowledge base | | **Tatiana** | Visual execution | Queries for style references when starting visual work | | **Jessica** | Art direction | Builds visual theses from pattern clusters | | **Ogilvy** | Copy | Searches for voice/tone references, messaging patterns | | **Anton** | Quality gate | Compares deliverables against approved references | | **Gerri** | Coordination | Attaches relevant references to project briefs | ### Use Cases **UC1 — Project Kickoff (Jessica/Tatiana)** > "We're starting visual identity for a fintech. Show me everything tagged `fintech` + `premium` + `minimal`. What patterns emerge? What's the visual thesis?" **UC2 — Copy Reference (Ogilvy)** > "Client needs confident B2B voice. Show me entries where copy critique mentions 'authority without arrogance'." **UC3 — Quality Check (Anton)** > "Compare this deliverable against our approved references for `editorial layout`. Does it meet the bar?" **UC4 — Trend Scouting (Julia)** > "Add 5 new entries from this week's design trends. Extract what's emerging vs. what's noise." **UC5 — Standards Evolution (Kitt)** > "This approved project output should become a reference. Add to knowledge base with extraction." --- ## 4. System Architecture ### 4.1 High-Level Flow ``` ┌─────────────────────────────────────────────────────────────┐ │ INPUTS │ │ │ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌─────────────┐ │ │ │ mymind │ │ Notion │ │ Manual │ │ CE Internal │ │ │ │ saves │ │ Inspo DB │ │ URL add │ │ Approved │ │ │ │ (Assaf) │ │ (team) │ │ (anyone) │ │ Work │ │ │ └────┬─────┘ └────┬─────┘ └────┬─────┘ └──────┬──────┘ │ │ │ │ │ │ │ └───────┼──────────────┼─────────────┼────────────────┼────────┘ │ │ │ │ ▼ ▼ ▼ ▼ ┌─────────────────────────────────────────────────────────────┐ │ PROCESSING │ │ │ │ ┌─────────────────────────────────────────────────────┐ │ │ │ 1. INGEST & CAPTURE │ │ │ │ Screenshot · Metadata · Tags · Source URL │ │ │ └──────────────────────┬──────────────────────────────┘ │ │ ▼ │ │ ┌─────────────────────────────────────────────────────┐ │ │ │ 2. AI EXTRACTION │ │ │ │ "What works" · "Why" · "Apply when" · "Avoid when" │ │ │ │ Copy critique · Visual principles · Pattern tags │ │ │ └──────────────────────┬──────────────────────────────┘ │ │ ▼ │ │ ┌─────────────────────────────────────────────────────┐ │ │ │ 3. HUMAN QUALIFICATION │ │ │ │ Assaf/Jessica review · Approve/reject/refine │ │ │ │ Score (1-10) · Add context AI missed │ │ │ └──────────────────────┬──────────────────────────────┘ │ │ ▼ │ │ ┌─────────────────────────────────────────────────────┐ │ │ │ 4. KNOWLEDGE BASE │ │ │ │ Structured JSON + Images + Extractions │ │ │ │ Searchable by: tag, principle, project type, │ │ │ │ visual style, copy pattern, industry │ │ │ └─────────────────────────────────────────────────────┘ │ │ │ └─────────────────────────────────────────────────────────────┘ │ │ ▼ ▼ ┌──────────────────────┐ ┌──────────────────────────────┐ │ HUMAN INTERFACE │ │ AGENT INTERFACE │ │ │ │ │ │ Web UI (browse, │ │ Query API: │ │ search, filter, │ │ - "fintech + premium" │ │ qualify, add) │ │ - "editorial layout refs" │ │ │ │ - "copy: authority voice" │ │ Quick-add bookmarkl.│ │ │ │ Mobile-friendly │ │ Returns: │ │ Bulk review queue │ │ - Matching entries │ │ │ │ - Extracted principles │ │ │ │ - Suggested visual thesis │ └──────────────────────┘ └──────────────────────────────┘ │ │ ▼ ▼ ┌─────────────────────────────────────────────────────────────┐ │ OUTPUT / USAGE │ │ │ │ ┌──────────────┐ ┌───────────────┐ ┌──────────────────┐ │ │ │ Project │ │ Visual Thesis │ │ Standards │ │ │ │ Briefs │ │ Documents │ │ Evolution │ │ │ │ │ │ │ │ │ │ │ │ Gerri attach-│ │ Jessica builds│ │ Proven patterns │ │ │ │ es relevant │ │ thesis from │ │ graduate from │ │ │ │ refs to each │ │ pattern │ │ taste board → │ │ │ │ new brief │ │ clusters │ │ STANDARDS.md │ │ │ └──────────────┘ └───────────────┘ └──────────────────┘ │ │ │ │ ┌──────────────┐ ┌───────────────┐ ┌──────────────────┐ │ │ │ Quality │ │ Copy │ │ Trend │ │ │ │ Benchmarks │ │ References │ │ Radar │ │ │ │ │ │ │ │ │ │ │ │ Anton checks │ │ Ogilvy pulls │ │ Julia cross-refs │ │ │ │ deliverables │ │ voice/tone │ │ new entries vs │ │ │ │ against refs │ │ patterns for │ │ existing to spot │ │ │ │ │ │ client work │ │ emerging trends │ │ │ └──────────────┘ └───────────────┘ └──────────────────┘ │ │ │ └─────────────────────────────────────────────────────────────┘ ``` ### 4.2 Agent Integration Flow ``` NEW PROJECT ARRIVES │ ▼ ┌───────────────┐ │ GERRI │ │ Receives │ │ brief │ └───────┬───────┘ │ Queries taste board: industry + style + context │ ▼ ┌───────────────┐ │ TASTE BOARD │──→ Returns 5-8 relevant refs │ Query API │ with extracted principles └───────┬───────┘ │ ┌────────────┼────────────┐ ▼ ▼ ▼ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ JESSICA │ │ TATIANA │ │ OGILVY │ │ │ │ │ │ │ │ Builds │ │ Uses │ │ Uses │ │ visual │ │ refs for │ │ copy │ │ thesis │ │ execut- │ │ patterns │ │ from │ │ ion │ │ for │ │ patterns │ │ style │ │ voice │ └────┬─────┘ └────┬─────┘ └────┬─────┘ │ │ │ └──────────────┼─────────────┘ ▼ ┌───────────────┐ │ ANTON │ │ Compares │ │ output vs │ │ references │ └───────┬───────┘ │ Pass ──┤── Fail → back to agent │ ▼ ┌───────────────┐ │ KITT │ │ Final review │ │ → Assaf │ └───────┬───────┘ │ If approved: entry goes BACK into taste board as internal reference (the loop) │ ▼ ┌───────────────┐ │ TASTE BOARD │ │ (evolved) │ └───────────────┘ ``` --- ## 5. Data Model ### Entry Schema ```json { "id": "uuid", "source": "mymind | notion | manual | internal", "// CAPTURE LAYER": "", "title": "Stripe Dashboard Redesign", "url": "https://stripe.com/dashboard", "screenshot": "screenshots/stripe-dash.webp", "tags": ["fintech", "dashboard", "data-viz", "minimal"], "type": "website | brand-identity | editorial | packaging | typography | illustration", "industry": "fintech", "created": "2026-02-15", "// EXTRACTION LAYER (AI-generated, human-verified)": "", "extraction": { "visual": { "what_works": "Dense data presented through generous whitespace and muted palette", "why": "Reduces cognitive load — user focuses on numbers, not chrome", "techniques": ["whitespace-as-hierarchy", "monochrome-data-viz", "type-scale-contrast"], "apply_when": "Data-heavy interfaces where trust matters", "avoid_when": "Consumer/playful contexts — feels too clinical" }, "copy": { "what_works": "Action labels are verbs, not nouns. 'View payments' not 'Payment history'", "why": "Reduces decision time — user knows what clicking does", "voice_pattern": "confident-minimal", "apply_when": "Any SaaS navigation or dashboard", "avoid_when": null }, "overall_principle": "Trust through restraint — every element earns its place" }, "// QUALIFICATION LAYER (human)": "", "qualification": { "score": 9, "qualified_by": "assaf", "qualified_at": "2026-02-16", "status": "approved | pending | rejected", "notes": "Reference-grade for any fintech project" }, "// INTERNAL WORK FIELDS (CE projects only)": "", "internal": { "project": "etoro-sow", "version": "v2-approved", "what_we_learned": "Client pushed back on bold type — but approved after seeing it in context" }, "// CONNECTIONS": "", "related_entries": ["uuid-1", "uuid-2"], "used_in_projects": ["etoro-sow", "flowrx-brand"], "graduated_to_standards": false } ``` --- ## 6. The Extraction Engine This is the core differentiator. Raw inspiration is noise. Extracted principles are signal. ### How Extraction Works ``` NEW ENTRY ADDED │ ▼ ┌──────────────────────────────────┐ │ STEP 1: AUTO-EXTRACTION (AI) │ │ │ │ Input: screenshot + URL + tags │ │ Output: │ │ - Visual analysis (what works) │ │ - Copy analysis (if applicable) │ │ - Pattern tags (techniques) │ │ - Suggested "apply when/avoid" │ │ - Related existing entries │ │ │ │ Agent: Tatiana (visual) + │ │ Ogilvy (copy) via Gerri │ └──────────────┬───────────────────┘ │ ▼ ┌──────────────────────────────────┐ │ STEP 2: HUMAN REVIEW QUEUE │ │ │ │ Assaf or Jessica reviews: │ │ - Is AI extraction accurate? │ │ - Score 1-10 │ │ - Add/edit principles │ │ - Approve / reject / refine │ │ │ │ UI: Card with screenshot + │ │ AI extraction + approve/edit │ └──────────────┬───────────────────┘ │ ▼ ┌──────────────────────────────────┐ │ STEP 3: KNOWLEDGE BASE ENTRY │ │ │ │ Only approved entries are │ │ queryable by the team. │ │ Rejected entries archived. │ │ Pending = visible but flagged. │ └──────────────────────────────────┘ ``` --- ## 7. Interfaces ### 7.1 Human Interface (Web UI) **Primary views:** | View | Purpose | Who | |------|---------|-----| | **Browse** | Visual grid of all approved entries, filterable by tag/type/industry/score | Everyone | | **Review Queue** | Cards pending qualification — AI extraction shown, human approves/edits | Assaf, Jessica | | **Search** | Semantic search: "premium fintech dashboards" returns ranked results | Everyone | | **Entry Detail** | Full card: screenshot, all extractions, usage history, related entries | Everyone | | **Add Entry** | URL input → auto-screenshot → auto-extraction → enters review queue | Everyone | | **Clusters** | Auto-grouped patterns: "these 12 entries share X principle" | Jessica, Kitt | | **Project View** | All references attached to a specific project | Gerri | **Quick-add methods:** - Web form (paste URL) - Notion integration (auto-sync from Inspiration DB) - mymind sync (periodic import) - Discord command: `/taste add [url]` → enters queue - Internal: approved deliverables auto-suggested for inclusion ### 7.2 Agent Interface (API) ``` // Query: Find references for a project brief POST /api/taste/query { "tags": ["fintech", "premium"], "type": "website", "limit": 8, "min_score": 7, "include_principles": true } // Response { "results": [...entries with extractions...], "patterns": [ "7 of 8 results use whitespace-as-hierarchy", "5 of 8 use monochrome palette with one accent", "Copy pattern: confident-minimal in 6 of 8" ], "suggested_thesis": "Premium fintech defaults to restraint — trust built through what's removed, not added." } // Add new entry POST /api/taste/add { "url": "https://example.com", "source": "manual", "tags": ["branding", "packaging"], "added_by": "julia" } // Log usage (tracks which refs influence which projects) POST /api/taste/log-usage { "entry_id": "uuid", "project": "flowrx-brand", "used_by": "tatiana", "how": "Color palette reference for hero section" } ``` --- ## 8. The Feedback Loop (Standards Evolution) ``` ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │ TASTE BOARD │────▶│ PROJECT │────▶│ APPROVED │ │ (references)│ │ (execution) │ │ (output) │ └──────────────┘ └──────────────┘ └──────┬───────┘ ▲ │ │ │ │ ┌──────────────────┐ │ │ │ STANDARDS.md │◀───────────┘ │ │ Updated with │ Proven patterns │ │ proven patterns │ graduate │ └────────┬─────────┘ │ │ └──────────────────┘ Best internal work re-enters as reference ``` **Rules for graduation:** 1. Pattern used successfully in 2+ projects → candidate for STANDARDS.md 2. Assaf explicitly approves a deliverable as "reference-grade" → enters taste board 3. EOD review (6pm daily) checks: "anything approved today that should become a standard?" --- ## 9. What We Already Have (Migration) | Asset | Count | Status | Migration Path | |-------|-------|--------|----------------| | mymind entries | 421 | JSON ready | Import → add extraction layer | | Notion Inspiration DB | 890 | API accessible | Sync → merge duplicates → extract | | Screenshots captured | ~2,645 | On disk | Map to entries, quality-check | | Notion cover images | 238 | Downloaded | Use as primary images where better | | VISUAL-STANDARDS.md | 1 | Active | Seed for "graduation" targets | | COPY-STANDARDS.md | 1 | Active | Seed for copy pattern extraction | **Migration priority:** Start with the 421 mymind entries (already structured). Add AI extraction. Build review queue. Then sync Notion. --- ## 10. Build Phases ### Phase 1: Foundation (Week 1) — ~15h - [ ] Data model finalized (this PRD → schema) - [ ] AI extraction pipeline (Tatiana + Ogilvy analyze entries) - [ ] Static site rebuild with search + filter + review queue - [ ] Import 421 mymind entries with auto-extraction - [ ] Basic agent query endpoint (JSON file, not API server) ### Phase 2: Integration (Week 2) — ~10h - [ ] Notion sync (cron pulls new entries from Inspiration DB) - [ ] Discord `/taste add` command - [ ] Gerri auto-attaches refs to project briefs - [ ] Usage tracking (which refs used in which projects) ### Phase 3: Intelligence (Week 3) — ~8h - [ ] Pattern clustering (auto-detect shared principles) - [ ] Visual thesis generation (query → synthesized recommendation) - [ ] Internal work pipeline (approved CE work → taste board) - [ ] Standards graduation tracking ### Phase 4: Refinement (Ongoing) - [ ] Semantic search (beyond tag matching) - [ ] Trend detection (Julia cross-refs new entries vs patterns) - [ ] Quality scoring recalibration based on usage data **Total estimated effort: ~33h across 3 weeks** --- ## 11. Success Metrics | Metric | Target | How to Measure | |--------|--------|----------------| | Time to first visual reference on new project | < 2 min | Agent query logs | | % of project briefs with attached references | > 80% | Gerri tracking | | Entries with completed extraction | > 90% | DB stats | | References actually used in project work | > 50 entries/month | Usage logs | | Standards graduated from taste board | 2+/month | STANDARDS.md git history | --- ## 12. Risks & Mitigations | Risk | Impact | Mitigation | |------|--------|------------| | Review queue bottleneck (Assaf's time) | Entries pile up unqualified | Batch review UI + auto-approve for score >8 from trusted sources | | AI extraction quality too low | Bad principles = bad work | Human review gate. Start conservative (all entries reviewed). | | Team doesn't query it | Tool becomes dead gallery | Embed in workflow: Gerri MUST attach refs to briefs | | Scope creep to full CMS | 3-month project | This PRD is the ceiling. JSON + static site. No backend servers. | --- ## 13. Technical Constraints - **No backend server.** JSON files + static site + Vercel. - **Agent queries via file read**, not HTTP API (agents read JSON, filter in-memory). - **Extraction runs as agent task**, not real-time pipeline. - **Images stored in repo** (compressed, <50MB total). - **Notion as source of truth** for entries; static JSON as read cache. --- ## 14. Open Questions for Assaf 1. **Review cadence:** Daily batch review, or as-they-come? How many minutes/day for qualification? 2. **Score threshold:** What score makes an entry "reference-grade" vs. "interesting but not directive"? 3. **Internal work inclusion:** All approved CE work, or only work you explicitly flag? 4. **Team write access:** Can anyone add entries, or only Assaf + Jessica + Julia? 5. **Copy extraction priority:** Equal to visual, or visual-first? --- *Using: zero-to-launch (MVP scoping) + strategic-build (LNO — this is Leverage work: compounds across every future project)*